Science/Tech/AI Futurism Thread

[aroneous]

Hello again, sorry for my brief hiatus (paper deadlines and such), I hope we are all doing well. For a while now I've been wanting to open a thread where we could discuss and speculate how specific developments in science, technology, and artificial intelligence could affect veganism and animal welfare as a whole, in both the near and longer terms. I think there is a whole lot to discuss here, especially considering how quickly science is advancing in the modern day. Of course I'm not expecting any of us to be experts on any particular scientific topic, but I think there are certainly evident technological trends in the past few years that even the most average layperson can pick up on, and some implications that can be drawn regarding how certain developments could relate to philosophical questions such as the reduction of animal suffering. Below I'll bring up three technologies that I think will be particularly relevant in the near future, but this is of course not an exhaustive list so feel free to point out anything else that I have neglected to mention here.

Robotics

One particularly exciting area of research that has the potential to change a lot, very quickly, is robotics. Robots already exist, of course, they have for several decades (and depending on your definition of a robot, perhaps even for millennia). But they don't feature very prominently in our daily lives at the moment. "Robots", in the sense of electrically powered machines that physically interact with the world through movement to manipulate real-world objects, are currently quite limited in scope. Their domain is primarily construction sites (excavators, cranes, forklifts, etc.), automotive assembly lines, the Martian surface, and, more recently, surgical operating rooms. We still don't see much of them in our daily lives, but I think that this may change very soon.

While we have a pretty good understanding of how robots work and how to make them do what we want them to do (according to kinematics equations) in an isolated environment, they still don't really operate in very close proximity with humans to, say automate everyday tasks like doing the laundry, preparing food, etc. You could say that these are more complex tasks that require much more "intelligence" than, say, the relatively fixed sequence of putting together components for a car, but that aside, robots still just aren't all that safe for humans to be around to begin with. Your typical robot is made up of several rigid metal segments joined together and actuated by powerful motors/hydraulics, with several tight corners (particularly at the joints) where a human body part that gets too close while the robot is in motion could get pinched, or worse. And a robot may cause blunt force trauma to a human if it moves too quickly when someone is nearby.

This safety issue is a tricky one to solve. If we wish to stick with conventional rigid-body robots, we will need to add a ton of sensors to them (for example, multiple sets of rapidly rotating stereoscopic cameras that give it a real-time 3D model of its current environment), as well as internal logic to instruct it to recognize and actively avoid contact with human body parts. You could imagine the first prototype of a safe rigid-body robot being, perhaps, some kind of auto-avoidant robotic arm that is impossible for a human to touch (except for, say, at the base where it is affixed to the wall) because it always knows exactly how to move out of the way when a human body part gets too close to it. Such technology could be feasible, even just using some base kinematic principles/PID control and without very heavy use of machine learning (other than for the 3D scene reconstruction and the ability to distinguish human hands from other objects). But still, putting together the full stack of technology needed for something like this will be a tremendous effort.

Another feasible route to safe robotics could be the emerging field of "soft robotics", that is, robots made of physically compliant components (such as rubber) that are much safer for humans and other animals to come into contact with. However, soft robots are a bit harder to reason about because of their more "floppy" nature, making it more difficult to program them to perform useful tasks. Purely soft robots must also rely on different means to produce the force needed to interact effectively with the world, since they cannot use conventional means such as motors, pistons, and servos, which require rigid components. However, to get the best of both worlds here, it could be possible to "merge" the two approaches -- namely, rigid robots encased in a soft body. As silly as this is to say, I think that the (fictional) robot "Baymax" from the Disney movie Big Hero 6 is a pretty accurate representation of what near-future human-centric robots might look like. Simply surrounding a rigid robotic skeleton with airbags seems to me like a pretty reliable way to make sure that critters can't get too close to the dangerous bits.

Brain Computer Interfaces

Another area of technology that I believe has the potential to significantly change the world in the coming years is the development of brain computer interfaces (BCIs). This sort of technology has existed for quite some time already, but there has been much more of a push in this direction in recent years from Elon Musk's company Neuralink. I'm not a big fan of Elon Musk or the fact that the company tests on animals -- IMO, they should have tested the installation of the implant on simulations/cadavers and then went directly into human trials, which would also have been FDA compliant, if a bit more costly on their end. But I do think that the BCI technology they are developing is very much future-relevant. In cases like this I try to separate the technology itself from the questionable means by which it was achieved, with the assurance that things would have been done more ethically in a vegan world.

It's also completely understandable that people are somewhat squeamish at the idea of having holes cut into their skull and wires inserted into their brain, and ideally the technology will advance to the point where it is minimally invasive (perhaps even non-invasive). And of course privacy concerns over the possibility of having your thoughts read and recorded by a third party. Personally though, the prospect of being able to control a computer with my thoughts is so alluring that I don't think I would have a problem getting such an implant, as long as there is some reasonable assurance of long-term safety. And regarding privacy, as long as there is some "off switch" that I can use when I don't want my thoughts to be read, that would be enough for me (though I'm not really a very privacy-oriented person to begin with).

Especially interesting is Neuralink's development of the "Blindsight" implant, which could enable enhanced human visual perception through direct stimulation of the visual cortex. Obviously such a device would have to be heavily invasive, and it may take quite some time for the technology to get to the point where they can implant a sufficient number of electrodes without critically damaging the integrity of the skull. But, beyond "curing" blindness by transmitting an encoded camera feed to the visual cortices of people with visual impairements, it could revolutionize the way ordinary people interact with computers, making computer monitors obselete by overlaying digital displays onto our normal visual perception.

Artificial Intelligence (Applied to Mathematical Proof)

It's of course impossible not also to address the recent hype around artificial intelligence, which right now seems to mostly be centered around large language models (LLMs) like ChatGPT and DeepSeek, and generative AI. Many people are convinced that LLMs in particular are "on the path to AGI", but I don't think that is really the case. They are certainly a very useful technology that are capable of automating a large class of mundane tasks, but I do not think that LLMs are "fundamentally useful" in terms of helping humans construct *new* knowledge. Language models can pick up on very subtle aspects of human reasoning that may even make them seem like "magic" at times, but at the end of the day they are just combining patterns that they have already seen. The limitation here arises not from the machine learning algorithms themselves, but rather the way in which they are being applied, namely by training them on large human-language corpora or image/video databases. While the results of this training are often *very* good, these systems still operate somewhat blindly, in that there is no way to know that what an AI system has produced is in fact "correct" according to some formally-defined standard.

In this respect, I believe that the next evolution of AI will come from integrations with systems of formal logical reasoning, in particular with tools known as "proof assistants", which are essentially programming languages that allow users to state and prove mathematical theorems, and have the system check that the proof is correct according to the theorem statement. Proof assistants essentially make it possible to encode and immortalize human logical understandings in a machine-checked formal language, and, crucially within the context of AI applications, provide a context in which an AI agent can "throw stuff at the wall and see what sticks", where something "sticking" is defined by a proof being accepted by the proof assistant. Though, it should be noted that a proof being correct does not necessarily mean that the property it shows was particularly useful to begin with. Since the space of provable propositions is literally infinite, a key challenge here will be to properly "nudge" the AI system into proving things in a generally useful direction.

My prediction is that the first effective applications of AI to proof assistants will be in the area of verified program synthesis (as many proof assistants have utilities for general programming), which is a smaller subdomain of mathematical proof that could narrow down the search space enough to make it possible to develop an AI system that is able to, say, generate a provably correct and efficient program that computes a certain function (e.g. a for sorting a list) given input and output specifications. This, I think, will be one of humanity's first "existential hacks" on the path to AGI. Such a tool could then possibly "bootstrap" the harder problem of general mathematical proof by enabling the generation of domain-specific programs for automated proving.

An AI tool for general mathematical proof could have the potential to "undercut" every other field of research -- at least, if this XKCD comic is anything to go on: https://xkcd.com/435/
There are indeed some problems that are outside the realm of mathematics itself (like the three body problem, for instance). But I think it may be possible that a substantial portion of the natural sciences can be based on some mathematical foundation. For instance, it may be possible for an AI system that operates purely based on logical deduction to make inferences regarding optimal medical treatments from a reading of the human genome. Or how to "rewrite" the genome in such a way that functionality is preserved while minimizing the probability of runaway cancerous growth resulting from mutations. I think that we might have a fairly straight shot here, but anyways, I may be wrong, this is all just (hopeful) speculation on my part.

Anyways, I realize that at this point I've done a lot of expositing my predictions for interesting potential upcoming tech and not a lot of discussing how they particularly tie back into veganism, but to avoid creating an impenetrable wall of text I think I'll cut myself off here and solicit some of your first thoughts on the matter. What are some emerging technologies that you are excited about, and how do you think they could impact the vegan movement/animal welfare in general?